Monday, March 27, 2017

Transcription

1 Monday, March 27, 2017 Topics for today Indexed branching Implementation of switch statement Reusable subroutines Indexed branching It turns out that arrays are useful in translating other language constructs, the C switch statement for example. Note that when we write BR LAB we mean BR LAB,i (the assembler lets us omit the,i for readability). From the table of Pep/9 instructions (e.g., inside textbook front cover) we see that it is also possible to write branches using indexing as in BR tab,x Q: What it the effect and why is it useful? A: The effect is to jump to whatever address is stored in Mem[tab+register X]. This means we can set up a table of addresses then branch quickly to any one of them based on the value in Register X. This is one way to implement a multi-way branch, e.g. a switch statement. In Fig. 6.40, Warford has an example of a switch statement translation here is a more general one: switch(w) { } case 13: case 16: < action 1 > break case 14: < action 2 > break case 11: < action 3 > break default: < action 4 > The Pep/9 corresponding to a switch statement has three components Comp 162 Notes Page 1 of 10 March 27, 2017

2 (1) implementations of the actions, with labels. Each action should end with a branch to take program control to whatever statement follows the end of the switch statement. This branch is the implementation of the break in the high-level example. (2) A table containing the labels of the action sections. Table[k] contains the label of the action we should take when the switch variable has value k or, more generally, k+constant. (3) A driver section to implement the branch. Here is an implementation of the example above. Part 1 - the actions action1:... code for whatever action 1 does goes here... br endsw this branch to stop us falling into action 2 action2:... other actions are similar... br endsw action3: br endsw action4: br endsw Part 2 - the table The.addrss directive in Pep/9 is similar to.word. Both have the effect of initializing a two-byte location with a value known at assembly time. Since we are storing addresses, we choose to use.addrss. In our example, if the switch variable has value 12 or 15 we go to the default case. swtab:.addrss action3 case 11.addrss action4 case 12.addrss action1 case 13.addrss action2 case 14.addrss action4 case 15.addrss action1 case 16 Comp 162 Notes Page 2 of 10 March 27, 2017

3 Part 3 - the driver section ldwx w,d the switch variable cpwx 11,i less than the lowest switch case? brlt action4 yes do the default case cpwx 16,i greater then the highest switch case? brgt action4 yes do the default case subx 11,i to get X in range of array indexes: 0.. aslx to make into offset for table of (2-byte) addresses br swtab,x jump to one of the actions endsw:... go here after each case Clearly this approach would not be a good one in all instances. What if the switch statement were switch(w) { case 1: < action 1 > case 100: < action 2 > case 1000: < action 3 > default: < action 4 > } The array in this case would be very large and almost all the entries would point to the default case. In this instance it would be better to implement the switch statement as a sequence of test-andbranch statements. A compiler will apply some internal check to the case labels to decide which of the two implementations to use. Indexed subroutine calls Note that CALL tab,x is also legal so we can set up a table of subroutine addresses and jump to one depending on the value on register X then urn to the instruction following the CALL. This might be a good way to implement a menu-driven program. Something like the following perhaps Comp 162 Notes Page 3 of 10 March 27, 2017

4 main: stro prompt,d output a prompt deci choice,d get user selection breq done assume entering zero means stop brlt main assume negative is invalid ldwx choice,d action number aslx convert to byte offset call options,x use table of subroutine addresses br main then get next choice done: stop options:.word 0 never executed corresponds to option 0 = stop.address sub1 sub1 to be called if user enters 1.address sub2 sub2 to be called if user enters 2 and so on sub1:. action corresponding to command 1. sub2:. action corresponding to command 2.. and so on. Comp 162 Notes Page 4 of 10 March 27, 2017

5 Reusable subroutines One of our motivations for using subroutines is that they can be made reusable, that is, they can be used in more than one program. What are some issues that might prevent us from cutting a subroutine from one program (source) and pasting into another (target)? A couple of problems are: (1) Labels that we use in the subroutine may clash with labels already in use in the target program. (2) The subroutine may interfere with the way that registers are being used in the target program, for example to control loops. Possible solutions (1) We can reduce the probability of label-clash if all the labels in a subroutine begin with a common prefix. However, rather than make this prefix a random string, we could base it on the name of the subroutine. 1 (2) We can save the values of registers A and X on entry to the subroutine and restore them on exit. A third characteristic of a reusable subroutine would be commentary describing parameters, results and any assumptions made about the environment in which the subroutine is used. We can add to our template, steps for saving and restoring registers Calling environment (1) allocate stack space for result (2) allocate stack space for actual parameters (3) push actual parameters on stack (4) do CALL (12) deallocate space for parameters (13) use then deallocate urned value Called environment (subroutine) (5) Allocate 4 bytes for and save A and X (6) allocate space for local variables (7) do action of the subroutine including.. (8) put urn value on stack (9) deallocate space for local variables (10) restore A and X and deallocate 4 bytes (11) do RET 1 Many assembly languages have a symbol (for example, a period or asterisk) that represents the current location. The symbol can be used in branch instructions avoiding the need for labels at all though making the code lees readable. BR *+6 BRLT.-12 Comp 162 Notes Page 5 of 10 March 27, 2017

6 Here is a modified version of our recursive function pstars to print N stars. pstars: subsp 4,i space for saved registers stwa 0,s saving A stwx 2,s saving X ldwa 6,s parameter breq pstars1 if zero, no printing ldbx '*',i stbx charout,d subsp 2,i for parameter of recursion suba 1,i stwa 0,s parameter is N-1 call pstars addsp 2,i deallocate parameter of recursion pstars1:ldwa 0,s restore A ldwx 2,s restore X addsp 4,i deallocate space for saved registers Our stack map now looks like Local Variables Saved A and X (4 bytes) Return Address (2 bytes) Parameters Space for urned value(s) Comp 162 Notes Page 6 of 10 March 27, 2017

7 Here is an example with comments describing usage and label conventions. bitcount Parameter N Result: number of 1's in the (two's complement) binary representation of N Label common prefix: bitco bitcount: subsp 4,i stwa 0,s saving registers on entry stwx 2,s ldwa 0,i register A counts the 1s ldwx 6,s register X holds the number being tested bitcotop: breq bitcofin zero means no more 1's brgt bitcolab skip if sign bit is zero adda 1,i count of 1's incremented because sign bit is 1 bitcolab: aslx shifting number brings next bit to sign position br bitcotop bitcofin: stwa 8,s count is urned as result ldwa 0,s restoring registers on exit ldwx 2,s addsp 4,i Reading Switch statements are covered in pages Comp 162 Notes Page 7 of 10 March 27, 2017

8 Review Questions 1. Regarding the switch statement, are there circumstances where the table element of the translation is redundant? 2. Suppose the switch variable can have values in the range but there are only 8 different actions. Devise an implementation that uses less space than a table of addresses. 3. How can we use indexed subroutine calls for a menu-driven system that has sub-menus? 4. Earlier in the semester we saw min3. Make min3, shown here again, reusable. min3: subsp 2,i local ldwa 4,s stwa 0,s local gets A cpwa 6,s compare with B brle skip B not smaller ldwa 6,s stwa 0,s smallest = B skip: cpwa 8,s compare with C brle skip2 C not smaller ldwa 8,s skip2:stwa 10,s urn smallest addsp 2,i 5. Here is an attempt to make mystro resuable. What is the error? mystro:subsp 4,i stwa 0,s stwx 2,s ldwa 0,i ldwx 0,i mystrol: ldba 2,sfx breq mystroex stba charout,d addx 1,i br mystrol mystroex: ldwx 0,s ldwa 2,s addsp 4,i Comp 162 Notes Page 8 of 10 March 27, 2017

9 Review Answers 1. Yes. If all the actions are the same number of bytes (e.g., B bytes) and switch value I maps to action I then the switch statement could be implemented as ldwx I,d < multiply register X by B> br Actions,x Where the actions are in consecutive memory locations with the first labeled Actions. 2. We could have an array BT of bytes (half the size of the normal address table). Entries in BT are integers If the switch value is k, we use BT[k] to select an action using a small table of 8 action addresses. 3. We could use a multi-dimensional array of subroutine addresses where each choice from the user takes us to the next dimension. Such an array may be jagged in that the number of choices in the sub-menus may be different. 4. Here is one possibility min3: subsp 4,i for saved registers stwa 0,s stwx 2,s subsp 2,i local ldwa 4,s stwa 0,s local gets A cpwa 6,s compare with B brle min3skip B not smaller ldwa 6,s stwa 0,s smallest = B min3skip:cpwa 8,s compare with C brle min3skp2 C not smaller ldwa 8,s min3skp2:stwa 10,s urn smallest addsp 2,i deallocate local ldwa 0,s ldwx 2,s addsp 4,i deallocate space for saved registers 5. The registers are restored from the wrong locations on the stack. Comp 162 Notes Page 9 of 10 March 27, 2017

10 Comp 162 Notes Page 10 of 10 March 27, 2017